A1083
Title: Predictive deep learning algorithm to analyze the effect of temperature fluctuation on suicide deaths
Authors: Ambreen Shafqat - Chonnam National University (Korea, South) [presenting]
Abstract: Background: Suicide is a significant public health issue in South Korea, where temperature fluctuations may impact suicide mortality rates. Limited research has explored temporal dynamics. Methods: Daily suicide mortality data were categorized by gender and age. Environmental variables included temperature fluctuation, dew point temperature, humidity, and weekday effects. Granger Causality tests identified optimal lags, and DLNM assessed nonlinear lagged associations. The LSTM model was trained in 73\% of the data and tested on 27\%. Results: Middle-aged males showed the highest suicide mortality, particularly in spring and summer. DLNM revealed significant lagged effects of temperature fluctuation at lag day 4 ($RR > 1.01$; 95\% CI: 1.01.018). LSTM outperformed in seasonal predictions, with the lowest error metrics in spring (0.12 vs. 0.14). LSTM provided better predictions at lag day 4, with a mean absolute error of 0.106. Dew point temperature and temperature fluctuation were the most influential predictors, with minimum error metrics (0.017 to 0.14). Conclusion: Environmental factors, particularly temperature fluctuation and dew point temperature, significantly influence suicide mortality in Seoul. DLNM captures lag-specific associations, while LSTM outperformed. Dew point temperature is the strongest predictor of seasonal suicide rates. These findings suggest the need for data-driven mental health strategies incorporating environmental risk assessments.
